A large number of computing apparatuses connected with network communication cables are used in a high performance computing, and an analytical computation in a parallel distributed processing by a plurality of computing apparatuses and a simulation computation by a plurality of computing apparatuses may be executed. For a large scale computing, a large number of computing apparatuses are required, and also, high performance computing apparatuses are needed.
A job control program is used for a control on jobs carried out by a large number of computing apparatuses. In the job control, a job is stored in a job queue, a necessary number of computing apparatuses for performing the job are secured, and the job is input from the job queue to the secured computing apparatuses. For the above-mentioned job control program, for example, an LSF provided by Platform Computing Corporation, a Network Queuing System (NQS) developed by Sterling Software Inc, and the like are proposed.
Also, it is proposed that a maximum temperature in a processing system provided with a plurality of processors is predicted on the basis of a detected temperature and stored information, for a plurality of deployment patterns indicating task allocations with respect to the plurality of processors, and a preferable task allocation to the processors is carried out.
With regard to job allocations to a plurality of computers, it is proposed that a temperature prediction based on a predicted value of power consumption and a heating value for the job allocations is carried out, and computers to which the job is allocated are decided on the basis of the result of the temperature prediction. Also, it is proposed that a temperature information management table is provided, a reference is made to the temperature information management table, a computer having a lowest temperature is searched for, and a job is allocated to the computer.
With regard to job allocation to a plurality of computing nodes, it is proposed that a value for determination is calculated for each inactive computing node on the basis of a power saving mode transition rate thereof and an average value of distances from communication targets thereto, and the job is allocated to an inactive computing node having the largest value for determination.
Related techniques are disclosed in Japanese Laid-open Patent Publication No. 2009-277022, Japanese Laid-open Patent Publication No. 2008-242614, Japanese Laid-open Patent Publication No. 2004-126968, and Japanese Laid-open Patent Publication No. 2009-223637.